The Prevalence Of Binge Eating Disorder And Associated Psychiatric And Substance Use Disorders In A Student Population In Kenya – Towards A Public Health Approach


 Background: Changing lifestyles in Kenya can lead to eating related behaviors and problems. The more severe problems are likely to manifest in clinical settings, but the majority and less severe forms will remain unrecognized. There is therefore the need to take a public health awareness approach to identify cases at community level and initiate appropriate intervention. This requires characterization of Eating Disorders (ED) and its associations in the local context. Our focus will be on the more common Binge Eating Disorder (BED). The overarching objective of this study is to generate Kenyan data on BED and fill a gap that exists not only in Kenya but Africa in general. The specific aims are: (1) To document the patterns and prevalence of different symptoms of BED in a student population whose age range represents a significant proportion of the population. (2) To determine associated psychiatric and substance use disorders (3) To determine independent predictors of BED. Method: We administered to a total of 9742 participants following tools: A researcher designed socio-demographic and economic indicators questionnaire; an instrument on DSM-IV diagnosis of BED and its various symptoms; instruments to determine DSM-IV psychiatric disorders, substance abuse, affectivity, psychosis and stress indicators. The participants were high school, college and university students in four out of the 47 counties in Kenya. We used descriptive and inferential analysis to determine prevalence and association of the different variables. The independent predictors of BED were generated from the generalized linear model (p<0.05). Results: We found a prevalence of 3.2% of BED and a wide range of BED symptoms varying from 8.1% to 19.0%. There were significant (p<0.05) associations between BED with various socio-demographic variables and psychiatric and substance use disorders. However, only some of these disorders were independent predictors of BED. Conclusion: Our findings on prevalence of BED and significant associations with various psychiatric disorders and substance use disorders are similar to those obtained in HICs using similar large scale samples in non-clinical populations. Economic status is not a predictor of BED. Our findings suggest a public health approach to awareness and management.

Using DSM-V increasing focus on large clinical sample and population studies: More recent studies using DSM-V have generally found higher prevalence of ED and same patterns of associated factors as found in earlier studies using DSM-IV. They have also used large population based samples. In an Australian population (Hay, Girosi, & Mond, 2015), a study was conducted and merged cross-sectional population survey data of adults aged over 15 years collected in 2008 and 2009 (n = 6041) on ED. They used the criteria based on DSM-V to make diagnosis using data already recorded in the questionnaire. By comparing DSM-IV and DSM-V on the same sample, the prevalence of bulimia nervosa and BED was reduced by nearly 50% on DSM-IV criteria compared with DSM-V criteria. On DSM-V, the prevalence of BED and sub-threshold BED were 5.6% − 6.9% (which was higher than other ED) with a mean age in the 4th to 5th decades compared with a mean age in the 4th decade for other ED. Thus, they found DSM-V expanded demographic distribution of ED to include adults.
In a report that summarizes clinical European studies and Eating Disorders published between 2015 and rst half 2016 (Keski-Rahkonen & Mustelin, 2016), BED prevalence ranged from < 1-4% with ED sub-thresholds of 2-3% of women and 0.3-0.7% men depending on ED. Over 70% of individual with different types of ED including BED reported co-morbidity disorders: Anxiety (> 40%), substance use (> 10%), somatic symptoms and increased risk of suicide. Risk factors included: parental psychiatric disorder, prenatal maternal stress, various family factors, childhood overweight and body dissatisfaction in adolescence.
They concluded that ED were relatively common, and were often overlooked (only about one third were detected by health care), although they were associated with high co-morbidity and serious health consequences.
In a sample of 36,309 adult participants in a nationally representative sample of adults in the USA, it was found that ED (anorexia nervosa, bulimia nervosa and binge eating disorder (BED)) were associated with lifetime DSM-V mood disorder, anxiety disorder, alcohol, drug use disorders and personality disorder and for BED following somatic disorders: diabetes, hypertension, high cholesterol and triglyceride (Udo & Grilo, 2019).
In a USA study using one of the largest samples of 71,712 randomly selected college students from 77 campuses found ED associations with suicidality (ideation and attempts).They found that students with the highest scores of ED had 11 times higher odds of attempting suicide as compared to two times in thresholds symptoms (Lipson & Sonneville, 2020). They concluded that ED should be a priority within campus suicide prevention efforts. In a systematic review, McDonald et al (McDonald, Rossell, & Phillipou, 2019) found that individuals with bipolar disorder had increased risk of developing ED compared to the general population, with ED lifetime or current ranging from 1.9-33.3%. They postulated that ED pathology may occur co-morbidly with bipolar disorder due to shared underlying pathophysiology, with potential implication intervention both pharmacological and psychotherapeutic but his requires more studies.
In a combined systematic review and meta-analysis, Michelle et al (Fornaro et al., 2020) found that in 36 studies that involved 15,084 primary bipolar patients and 11 studies that involved 15,146 people with primary ED found that BED occurred in 12.5% of bipolar disorders. Overall, bipolar patients with ED were more likely to be females compared with non-ED controls (Fornaro et al., 2020).
The prevalence of co-morbidity PTSD in patients with ED ranges from 9 to 24%, and that co-morbidity PTSD is associated with more severe ED symptoms (Rijkers, Schoorl, van Hoeken, & Hoek, 2019).
In a meta-analysis study on the rates of co-morbid obsessive compulsive disorder in ED using a binary random effects model, Mandelli et al (Mandelli, Draghetti, Albert, De Ronchi, & Atti, 2020) found that globally, 18% and 15% of all patients with ED had a lifetime and current co-morbidity with OCD in clinical populations respectively, using DSM-V criteria. They concluded that OCD co-morbidity of ED affects almost 20% of patients in cross-sectional observations and up to nearly 40% in prospective follow up studies and therefore necessary to consider OCD in order to arrive at a better diagnostic and prognostic arriving and targeted treatment.
From the available literature, nearly all the studies have been reported from High Income Countries (HIC) Western countries and one from Iran. In particular, no studies have been done in Kenya relating to community prevalence of ED, the different symptoms of BED, associated psychiatric conditions including psychosis and SUD and what independently predicts BED. The overall objective of this study is to ll that gap not only for Kenya but for Africa in general.
The speci c aims are: 1. To document the patterns and prevalence of different symptoms of BED in a student population that represents signi cant properties of Kenyan population.
timetables each day of the week. The high school students were directed to speci c public meeting areas for assessment with the help of the local community leaders. Participants were only included in the study if they met the requirements i.e. were in high school, college or university, had voluntarily agreed to participate in the study. Informed consent was obtained from participants aged 18 + and from parents of participants that were below the age of 18 and also assented. The age range for high school students is 14-18 years. For college/university the age range starts 19-25 for most students but also have late entry students. High school, college and university students in Kenya are all uent in English -English is a national language and the o cial language for all communications and medium of learning. All participants, regardless of their age and provided they had been o cially registered as students were included and treated as students. A self-administered questionnaire was used to collect data from participants. A total of 9,742 participants from different years of study and courses were recruited for the study. As part of preparation for this study, we discussed with institutions on the need to incorporate mental health in their institutional health services for their students in case the awareness of mental disorders during the study prompted students to seek for mental health services. As for the high school students, we had trained staff at the local health center facility on the WHO mhGAP-IG. We also informed and directed the participants where to seek help at their institutions and community levels in case they needed any help. All the participants who were approached participated.
Tools 1. Socio-demographic characteristics: A researcher-designed questionnaire was used to get the socio-demographics information of the respondents. Sociodemographic variables included age, gender, highest level of education, marital status and birth order.
2. Economic Indicators: The respondents completed questions regarding household items, water source, toilet type and cooking method. These were used to estimate socio-economic status by creation of wealth index (Smits & Steendijk, 2015). The wealth index used is based on the World Bank Recommendation for LMICs (Smits & Steendijk, 2015) and has been adopted by the Kenya Government for use in Kenya. It is classi ed into ve sections, quintile 1-5 with quintile 1 representing the lowest level of wealth and 5 the highest level.
3. Psychiatric conditions: The psychiatric diagnostic screening questionnaire (PDSQ) was used to assess psychiatric conditions of the respondents. It consists of 126 questions assessing the symptoms of 13 DSM-IV Axis 1 disorders: mood disorders (major depressive disorder [MDD]); anxiety disorders (panic disorder, agoraphobia, PTSD, obsessive-compulsive disorder, generalized anxiety disorder [GAD], and social phobia); substance use disorders (alcohol abuse/dependence and drug abuse/dependence); and somatoform disorders (somatization disorder and hypochondriasis). In addition, it has a 6-item psychosis screen. For each psychiatric disorder, there are several questions which are computed to arrive at a diagnosis. The disorders chosen for coverage were selected because they are the most prevalent in epidemiological surveys of the community (Kessler et al., 1994;Robins, 1991) and the most frequently reported in large clinical samples (Koenigsberg, Kaplan, Gilmore, & Cooper, 1985;Mezzich, Fabrega, Coffman, & Haley, 1989;Shear et al., 2000). In a validity study in which 994 psychiatric outpatients completed the scale (Zimmerman & Mattia, 2001), the 13 PDSQ subscales demonstrated good to excellent levels of internal consistency. Cronbach's α was greater than .80 for 12 of the 13 subscales, and the mean of the α coe cients was .86.
Test-retest reliability was examined in the 185 subjects who completed the PDSQ 2 times less than a week apart. Test-retest reliability coe cients were greater than 0.80 for 9 subscales, and the mean of the test-retest correlation coe cients was 0.83. The convergent and discriminant validity of the PDSQ subscales (Campbell & Fiske, 1959) was examined in 361 patients who completed a package of questionnaires at home less than a week after completing the PDSQ. The last six questions from PDSQ on major depressive episode domain are used to measure suicidal ideation classi ed as follows: frequently think of dying in passive ways like going to sleep and not waking up, wishing to be dead, thinking you were better off dead, having thoughts of suicide, seriously considering taking life, and thinking about speci c ways of taking your life. Further, PDSQ has 10 questions on BED each question asking about speci c symptoms of BED. The 10 questions are computed to make a diagnosis of BED. The questions were coded as No or Yes with No having a value of zero and Yes having a value of one.
4. Psychosis, affectivity: The Washington Early Recognition Center Affectivity and Psychosis (WERCAP) screen was used to quantitatively assess psychosis-risk symptoms and bipolar-risk symptoms (affectivity) based on the frequency of symptoms and their effects on functioning (Mamah, Owoso, She eld, & Bayer, 2014). It has high test-retest reliability and validity, with affectivity of sensitivity of .91, speci city of .71, psychosis sensitivity .88 and speci city of .82 (Mamah et al., 2014).

Drug and substance:
The WHO's Alcohol, Smoking, and Substance Involvement Screening Test (ASSIST) (Humeniuk et al., 2010) was used only to determine the prevalence of different types of substance use on a Yes or No dimension.

Data management and Analysis
The coded data was checked, cleaned and exported into Statistical Package for the Social Sciences (SPSS) version 23.0 for analysis. Basic descriptive statistics were carried out to estimate the prevalence of Bulimia/binge eating disorder as well as the participant's socio-demographic and socio-economic characteristics. Scores were grouped into two: those with the eating disorder and those without the eating disorder. Estimation of univariate associations between Bulimia/binge eating disorder and other variables was carried out by tting bivariate logistic regression that was also used to identify confounding factors. Variables with p-value less than 0.05 were then tted into generalized linear model, with logit as the link, to identify independent predictors of Bulimia/Binge eating disorder. The strength and signi cance of the association between the variables and Bulimia/Binge eating disorder was assessed by calculating the adjusted odds ratio with a 95% con dence level. Correlation analysis was also carried out between Bulimia/Binge eating disorder and the psychiatric conditions. All the tests carried out were two-sided with a set P-value of less than 0.05 (p < 0.05). Table 1 presents the results of socio-demographic characteristics of the respondents. Males (53.5%) were more than females. Mean age was 21.4, median 21.3 (range 15-43) years. Majority of the respondents were single (93.4%), mainly from university (68.6%), most (56.9%) were either rst or second born in their families. The wealth index was evenly distributed among the rst 5 quintiles with the fth quintile having the lowest proportion (16.6%).  Figure 1 presents a pictorial representation in descending order of BED symptoms using 10 PDSQ items used to screen for BED.

Associated Factors
i. Socio-demographic factors and wealth index Table 2 summarizes the socio-demographic factors associated with BED at bivariate level. The younger age group, representing high school students were 1.39 times more likely to have Bulimia/Binge Eating Disorder as compared to those in the university (p < 0.05). No other socio demographic factors or level of wealth index was associated with BED.  Table 3 presents the results on the psychiatric disorders associated with BED at bivariate level. There was signi cant association between all psychiatric disorders and BED (p < 0.001). The correlation between BED and psychiatric disorders was highly signi cant (p < 0.001) for all psychiatric disorders (Table 4). Note; *P < 0.05; **P < 0.01, ***P < 0.001 iii. Drug and substance use Table 5 presents the results on drug/substances associated with BED at bivariate level. There was signi cant association between all the substances and BED except for alcohol (p < 0.05). The prevalence of SUD in BED varied from 3.7-7.6%. Alcohol had the lowest prevalence of 3.7%. Tables 6 summarizes the independent predictors of BED using AOR analysis. Major depressive disorder, OCD, panic disorder, psychosis, agoraphobia, drug abuse, generalized anxiety disorder, suicidality and WERCAP stress were the leading independent predictors of BED.

Discussion
We present the rst African reported non-clinical population epidemiological patterns of different symptoms in BED to be reported speci cally in a wide range of students with a large sample as opposed to clinical epidemiological patterns that are based on clinical populations. To the best of our knowledge, it is one of the two reported studies on college students globally (the other one was done in USA college students) and one of the few large sample non-clinical populations studies reported in global literature. It is also one of the two large sample reported globally to investigate ED across a wide age range, starting from 15 years to adulthood reported by Phillipa et al (Hay et al., 2015) in Australia.

Response rate
The 100% response rate in our study is not unique and is common in Kenyan community-based mental health related and student surveys (Ndetei, Khasakhala, Mutiso, Ongecha-Owuor, & Kokonya, 2009. Students and parents place immense value on such surveys, as education is regarded as the best investment, with the highest potential to propel students into successful futures and help them and their families escape from poverty. However, we point out that we approached colleges and university students at speci c time points as a "captive" group, which may have contributed to the high response rate. The high school students voluntarily came to the community centers, which means they were already motivated to participate.

The Socio-Demographics
Males were more than the females for the following two plausible reasons. Most of the college students were in technical colleges that are patronized by male than female gender. Secondly, the data collection for high school students was done when the schools were closed. Boys more than girls were more likely to be allowed to go to the data collection site. The wide range in age is easily explained by late age entry to schools, colleges and university. Since most of the participants were students, it is not surprising that most were single. Other types of marriages were come we stay arrangement common in college and university students as one of the ways of sharing accommodation costs.
Kenyan family structure is increasingly getting smaller, explaining 1st and 2nd birth order (Kenya National Bureau of Statistics (KNBS), 2019). We did not access as many high school students as we did in the captive groups that we found in college and university students since the data was collected when the schools were closed. There was an equitable distribution of wealth index but it was less for quintile 5 as compared with the rest. This is most likely a re ection of wealthy parents taking children to private institutions which were not represented in our study. Our ndings suggest that BED is not a re ection of socioeconomic status, even when this socio-economic status may re ect relative availability of food.

Prevalence of BED
The overall prevalence of BED using DSM-IV criteria (3.2%) found in our study is much higher than the "estimated" general population lifetime prevalence of BED of 0.3-1.6% reported by Swanson et al (Swanson et al., 2011) using DSM-IV criteria but within the range of 0.7-4.3% reported by American Psychiatric Association (APA)(American Psychiatric Association, 2013), using the DSM-V. If we apply the nding of Philippa et al (Hay et al., 2015) in Australia that DSM-V expanded ED by 50% over DSM-IV, which was used in this study, then the 3.2% found in this study could be increased by 50% to a prevalence of 4.8%. This is just above the 4.3% upper level reported by APA using DSM-V criteria (American Psychiatric Association, 2013). These ndings are essentially identical. Even much higher are the prevalence of individual BED symptoms. Our sample was drawn from student populations as opposed to the general population. It is possible student population in institutions are different from general population and therefore cannot extrapolate this to the general population. However, in favor of plausible extrapolation to the general populations is that these students were drawn from the general population. Given the pyramid structure of the Kenyan population (Kenya National Bureau of Statistics (KNBS), 2019), these students represent a signi cant proportion of the total Kenyan population.
We nd our results surprising that BED could be 3.2% and possibly 4.8% if we had used DSM-V and similar to WPA ndings, coming from LMIC setting, often associated with limited resources and often reported malnutrition or shortage of food (Grace, Davenport, Funk, & Lerner, 2012).
We could not nd any study anywhere that reported the prevalence of individual symptoms to compare with our ndings of 19.0-8.1% of the various items.
These prevalence of individual symptoms have implications on public approach to BED and to some extent clinical practice for the purpose of creating awareness in the public and for clinical practices.

BED and psychiatric Morbidity
The high psychiatric co-morbidity with BED found in our study is similar to ndings in many other countries, mainly high income countries and also mainly from clinical and non-clinical populations cited under our Background, whether using DSM-IV or DSM-V criteria.
Our study therefore con rms the ndings in our literature review (Background) that BED is associated with various psychiatric disorders. Our study reports an association not reported elsewhere, that is ED is associated with stress. However, our study did not explore the cause-effect relationship between different psychiatric disorders and BED.
Our ndings are similar to ndings in High Income Countries on the association between BED and substance use disorder (SUD) whether using DSM-IV or DSM-V criteria for BED. However, direct comparison on the prevalence of SUD in BED is not possible because of the different methods used to determine SUD in BED and whether lifetime or current use. This could be the explanation for the nding in our study of an average of 3.1-7.6% which is lower than the lifetime reported prevalence of 30% − 70% lifetime use in BED reported elsewhere in HICs and highlighted in our literature review, by Bellodi et al 2001and Herpetz et all 2001(Bellodi et al., 2001Herpertz-Dahlmann et al., 2001). Our ndings therefore concur with the ndings in HIC that there is an association between SUD and BED. This difference between our ndings and those from HICs could be that the drinking patterns are different in the different research participants.
Our study could not however establish the actual relationship between these associations -whether causal or casual and how the SUD is related to comorbidity with ED. Qualitative and biomarker studies are required to elucidate the nature of this association even in the Kenyan setting. Our ndings suggest the need for routine screening of BED in clinical cases of BED and vice-versa.

Independent Indicators
It is noteworthy that no socio-demographic variable was associated with BED not even assumed age, given that high school students are younger than the rest. It is also noteworthy that wealth index of the family was not an independent predictor of BED. These ndings suggest that BED occur in all sociodemographic variables reported in our study and also that BED occurs across the economic spectrum. The implication for this is BED is not a condition necessarily associated with a uence, at least in our study participants. Further, the fact that we found same prevalence in economically different countries with different cultural backgrounds would suggest cultural context has no bearing on expression of BED symptoms although levels of awareness may be different. Western studies cited in our literature have indicated that ED may go unnoticed even in clinical settings. Though both alcohol disorder and SUD were more in BED than in Non-ED, it was SUD that achieved signi cant levels. Most of the independent predictors were major depression and various types of anxiety disorders and increased stress levels. It is therefore not surprising that sedatives that are essentially anxiolytics were signi cantly associated with ED.
These should be screened for and factored in any management of persons with ED.

Implications of our ndings
There is need for public health awareness that BED is found in Kenya. The fact that we found a prevalence of 3.2% (and possibly 4.8% if we had used DSM-V criteria) in non-clinical student population means that the prevalence is higher in clinical population awareness at both clinical and community levels. This is necessary for clinicians, public health practitioners, family members and potential sufferers of ED. It is possible that students presenting with psychiatric disorders have existing BED that is not recognized and therefore not addressed. The youth may not volunteer this information unless prompted and even so may deny the existence of such a problem. A DSM-V based screener may be an easier and acceptable way to nd out the existence of BED symptoms.

Strengths of the study
This study had a large sample of 9742 treatment naïve non clinical populations of students in four out of the 47 counties. The college and university students were from across the country because of the nationally centralized admissions to colleges and universities.
Limitations of the study A major limitation of this study is that we used DSM-IV and not DSM-V leading to lesser prevalence. Because of the large sample it was not realistic to use researcher administered interview approaches. However, in population based surveys it is not uncommon to use self-administered screening instruments with the double edged risks for under and or over reporting. However, the use of DSM-IV had no effect on the various association.
Another main limitation is that this was a study on students as opposed to the general population. In mitigation, most youth in the communities belong either to the school, college and university going groups. They therefore represent the general population within the age groups studied. They also reported a signi cant proportion of the Kenyan population, given the pyramid population structured in Kenya. Although there were late age entry students, the contribution of age was determined at analysis level and found that age was not a predictive factor. A further limitation of our study was we did not include AN, which however is a lessor prevalent aspect of the overall Eating Disorder. However, our study with a wide age range is similar to other studies reported in the literature.

Conclusion
Subject to the limitations above, we have achieved our overall objective and the speci c aims. In particular we have established the prevalence of BED, associated psychiatric conditions and SUD, the independent predictors of BED. On the basis of all these, we are able to suggest potential approach on possible interventions through public and service providers' awareness.

Future Studies
This study was essentially an exploratory cross-sectional study. However, our ndings point out to areas that need further studies and in particular the nature of the association between ED, psychiatric disorders and SUD. Mixed methods i.e. qualitative and quantitative including biomarker and a longitudinal approach would determine the nature of the associations. Prevalence studies of AN and how they co-morbid with BED are recommended.
A peculiar aspect of this study is not so much what value it adds to studies in High Income Countries, but that LMIC's constitute the largest proportion in addition to greater burden of mental disorder of global population, and therefore on important contribution to the global data base. Further, the environment which varies greatly across different settings, interact with biological factors to determine development and behavioral outcomes, and therefore important to see how prevalence and associations compares between HIC and LMIC.

Declarations
Page 13/15 Ethics approval and consent to participate: Ethical approval was granted by the Maseno University Ethics Review Board in Kenya (IRB number MSU/DRPI/MUERC/00344/16).
Informed consent: Informed consent was obtained from participants aged 18 + and from parents of participants who were below the age of 18. Participants below 18 years signed assent forms.

Consent for publication:
Not applicable Competing interests: None Guidelines All the guidelines have been followed and adhered to as per the journal requirements. Funding: Africa Mental Health Research and Training Foundation (AMHRTF) in-house support for this particular set of data collection and analysis. The Foundation did not in uence researchers work independently.
Authors' contributions: VNM -oversight on data collection; drafting of the paper; DMN -conceptualization and oversight of the study; drafting of the paper; ENM -literature review;; RKdata analysis; LO-data analysis; FK-drafting ; MM-literature review ; CM -Ethical conduct of the study and data management; DM -conceptualization of the study.